A data catalog tells you what exists. Observability tells you whether you can trust it this morning.
Those are not the same product. Teams that confuse them buy a map of the warehouse and still ship decisions on stale or broken tables.
If the catalog cannot answer "is this fresh, who owns it, and what depends on it," you did not buy trust. You bought searchable inventory.
Inventory is not a signal
A catalog is an inventory. It holds names, descriptions, tags, and often a lineage graph.
That is useful. People need to find tables. New hires need a starting point. Auditors need a list.
None of that proves the data is fit for use today.
Searchable bad data is still bad data. A polished entry for a fact table that missed last night's load does not help anyone. It helps them find the wrong number faster.
Observability is the operating signal:
- Freshness
- Volume and schema change
- Failed jobs
- Lineage you can trust when something breaks
- An owner on the path who can act
Buy that signal for the few tables that run the business. Then expand the catalog. Not the other way around.
What I see when the catalog comes first
The pattern is common.
Leadership funds a catalog. The team connects every warehouse schema. Weeks go into descriptions and tags. The demo looks clean.
Adoption stalls after the first month because people still open Slack to ask if yesterday's numbers are good.
Why. The catalog answered "where is the table." It did not answer "can I use it."
I have watched teams spend a quarter on coverage while the revenue fact still had no freshness check and no named owner. The catalog entry looked complete. The operating path was empty.
That is not a tooling miss. That is a sequencing miss.
A working example from the critical path
Here is a sequence I use as a fractional CDO before any catalog rollout.
Pick three tables that already drive a weekly decision. Write them down. Common picks include:
- Revenue
- Orders
- Claims
- Inventory
Whatever your business actually steers on.
For each table, require four things before you celebrate inventory:
- A freshness rule with a threshold and an alert that reaches a person.
- A named owner on payroll who can fix a miss the same day.
- Downstream consumers listed in plain language, not only in a graph nobody opens.
- A simple "do not use if" note for the people who trust that table.
Then run that for two weeks. Read the alerts. Fix the false ones. Fill the owner gaps.
Only after those three paths are quiet do I green-light a wider catalog push.
At one company I supported, the marketing dashboard had a polished catalog page. The entry looked complete. Descriptions, tags, and lineage were filled in.
The upstream load had been late three mornings in a row. The chart on the dashboard still rendered. Nothing on the page said the feed was late. Marketing treated the numbers as current because the catalog looked done and the chart looked normal.
Nobody saw an alert. They only saw a chart that still painted.
We did not add more catalog fields. We put a freshness check on the feed table, named an owner, and wired a miss to Slack. The next late load reached a person before the meeting started.
The catalog page stayed. Trust moved because the signal arrived first.
That is the order. Signal first. Inventory second.
Freshness, lineage, and owners are the bar
Three pieces must land together.
Freshness. Someone knows if the table is late. A SLA or a simple "should land by 6 AM CT" rule is enough to start. Silence is not a status.
Lineage people trust. A graph that is wrong is worse than no graph. Start with the one or two hops that matter for the critical tables. Keep it true. Expand later.
Owners on the critical paths. A tag that says "Data Team" is not an owner. A person who can change the job and answer at 7 AM is an owner.
If any of those three is missing, the catalog is a directory. It is not observability.
What to buy before the inventory SKU
Ask vendors and internal platform teams for the operating pieces first.
- Alerts on freshness and failed loads for the tables that move money or risk.
- A short owner map for those tables, reviewed monthly.
- Lineage that matches the real pipeline for those same paths.
- A weekly 30-minute review of misses, not a quarterly catalog cleanup.
A catalog can sit on top of that later. It will then point people to assets that already have a pulse.
If the budget only funds the inventory layer, you will get a searchable warehouse and the same Slack questions you have today.
What I would do this week
List the five datasets your leadership already argues about. Do not open a catalog RFP yet.
For each one, write the owner, the freshness rule, and the last time anyone checked a miss. Fill the blanks. Wire one alert that reaches a human.
When those five are honest, you have the start of observability. The catalog can wait until the signal is real.